Search

The generation of artificial data based on existing observations, known as data augmentation, is a technique used in machine learning to improve model accuracy, generalisation, and to control overfitting. Augmentor is a software package, available in both Python and Julia versions, that provides a high level API for the expansion of image data using a stochastic, pipeline-based approach which effectively allows for images to be sampled from a distribution of augmented images at runtime. Augmentor provides methods for most standard augmentation practices as well as several advanced features such as label-preserving, randomised elastic distortions, and provides many helper functions for typical augmentation tasks used in machine learning.

In view of the high number of deaths and complication rates of major surgical procedures worldwide, surgical safety is described as a substantial global public-health concern. Naturally, patient safety has become an international priority. The increasing amount of electronically available clinical documents holds great potential for the computational analysis of large repositories. However, most of this data is in textual form and the clinical domain is a challenging field for the appliance of natural language processing. This is particularly the case if you deal with a language other than English, due to the little attention from the international research community. In this project, we are concerned with the utilization of a Germanspeaking operative report repository for the purpose of risk management and patient safety research. In this particular paper we focus on the description of our information retrieval approach. We investigated the thought process of a domain expert in order to derive his information of interest and describe a facet-based way to navigate this kind of information in the form of extracted phrases. Initial results and feedback has been very promising, but a formal evaluation is still missing.

Heart rate variability is the variation of the time interval between consecutive heartbeats. Entropy is a commonly used tool to describe the regularity of data sets. Entropy functions are defined using multiple parameters, the selection of which is controversial and depends on the intended purpose. This study describes the results of tests conducted to support parameter selection, towards the goal of enabling further biomarker discovery. This study deals with approximate, sample, fuzzy, and fuzzy measure entropies. All data were obtained from PhysioNet, a free-access, on-line archive of physiological signals, and represent various medical conditions. Five tests were defined and conducted to examine the influence of: varying the threshold value r (as multiples of the sample standard deviation σ, or the entropy-maximizing rChon), the data length N, the weighting factors n for fuzzy and fuzzy measure entropies, and the thresholds r F and r L for fuzzy measure entropy. The results were tested for normality using Lilliefors' composite goodness-of-fit test. Consequently, the p-value was calculated with either a two sample t-test or a Wilcoxon rank sum test. The first test shows a cross-over of entropy values with regard to a change of r. Thus, a clear statement that a higher entropy corresponds to a high irregularity is not possible, but is rather an indicator of differences in regularity. N should be at least 200 data points for r = 0.2 σ and should even exceed a length of 1000 for r = rChon. The results for the weighting parameters n for the fuzzy membership function show different behavior when coupled with different r values, therefore the weighting parameters have been chosen independently for the different threshold values. The tests concerning r F and r L showed that there is no optimal choice, but r = r F = r L is reasonable with r = rChon or r = 0.2σ. Some of the tests showed a dependency of the test significance on the data at hand. Nevertheless, as the medical conditions are unknown beforehand, compromises had to be made. Optimal parameter combinations are suggested for the methods considered. Yet, due to the high number of potential parameter combinations, further investigations of entropy for heart rate variability data will be necessary.

Professionals in the Life Sciences are faced with increasing masses of complex data sets. Very few data is structured, where traditional information retrieval methods work perfectly. A large portion of data is weakly structured; however, the majority falls into the category of unstructured data. To discover previously unknown knowledge from this data, we need advanced and novel methods to deal with the data from two aspects: time (e.g. information entropy) and space (e.g. computational topology). In this paper we show some examples of biomedical data and discuss a taxonomy of data with the specifics on medical data sets.

Part 1: Cross-Domain Conference and Workshop on Multidisciplinary Research and Practice for Information Systems (CD-ARES 2013); International audience; Many research problems are extremely complex, making interdisciplinary knowledge a necessity; consequently cooperative work in mixed teams is a common and increasing research procedure. In this paper, we evaluated information-theoretic network measures on publication networks. For the experiments described in this paper we used the network of excellence from the RWTH Aachen University, described in [1]. Those measures can be understood as graph complexity measures, which evaluate the structural complexity based on the corresponding concept. We see that it is challenging to generalize such results towards different measures as every measure captures structural information differently and, hence, leads to a different entropy value. This calls for exploring the structural interpretation of a graph measure [2] which has been a challenging problem.

Medical professionals are confronted with a flood of big data most of it containing unstructured information. Such unstructured information is the subset of information, where the information itself describes parts of what constitutes as significant within it, or in other words - structure and information are not completely separable. The best example for such unstructured information is text. For many years, text mining has been an essential area of medical informatics. Although text can easily be created by medical professionals, the support of automatic analyses for knowledge discovery is extremely difficult. We follow the definition that knowledge consists of a set of hypotheses, and knowledge discovery is the process of finding or generating new hypotheses by medical professionals with the aim of getting insight into the data. In this paper we present some lessons learned of ICA for dermatological knowledge discovery, for the first time. We follow the HCI-KDD approach, i.e. with the human expert in the loop matching the best of two worlds: human intelligence with computational intelligence.

Information entropy as a universal and fascinating statistical concept is helpful for numerous problems in the computational sciences. Approximate entropy (ApEn), introduced by Pincus (1991), can classify complex data in diverse settings. The capability to measure complexity from a relatively small amount of data holds promise for applications of ApEn in a variety of contexts. In this work we apply ApEn to ECG data. The data was acquired through an experiment to evaluate human concentration from 26 individuals. The challenge is to gain knowledge with only small ApEn windows while avoiding modeling artifacts. Our central hypothesis is that for intra subject information (e.g. tendencies, fluctuations) the ApEn window size can be significantly smaller than for inter subject classification. For that purpose we propose the term truthfulness to complement the statistical validity of a distribution, and show how truthfulness is able to establish trust in their local properties.